s4_e0
done · peak unseen_mismatch 92.6
α schedule
α 4.0 ×50 α 3.8261 ×50 α 3.6522 ×50 α 3.4783 ×50 α 3.3043 ×50 α 3.1304 ×50 α 2.9565 ×50 α 2.7826 ×50 α 2.6087 ×50 α 2.4348 ×50 α 2.2609 ×50 α 2.087 ×50 α 1.913 ×50 α 1.7391 ×50 α 1.5652 ×50 α 1.3913 ×50 α 1.2174 ×50 α 1.0435 ×50 α 0.8696 ×50 α 0.6957 ×50 α 0.5217 ×50 α 0.3478 ×50 α 0.1739 ×50 α 0.0 ×50
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.0
seen_* probed at reference α = 1.4 · unseen_* = held-out pairings, uniform (α-independent) · 1000 pairs per condition
| iteration | seen_match | seen_mismatch | unseen_match | unseen_mismatch |
|---|---|---|---|---|
| 50 | 51.7 ± 1.9 | 49.6 ± 1.0 | 49.9 ± 2.1 | 50.5 ± 1.2 |
| 100 | 58.6 ± 8.2 | 56.6 ± 6.6 | 54.3 ± 5.8 | 52.6 ± 4.2 |
| 150 | 78.0 ± 6.7 | 77.1 ± 6.0 | 65.7 ± 7.8 | 62.4 ± 7.1 |
| 200 | 84.2 ± 1.7 | 82.5 ± 1.5 | 68.3 ± 3.3 | 65.3 ± 3.1 |
| 250 | 87.3 ± 1.9 | 85.5 ± 1.5 | 69.8 ± 4.5 | 67.9 ± 3.3 |
| 300 | 90.5 ± 1.8 | 89.4 ± 1.5 | 74.9 ± 3.3 | 71.4 ± 4.6 |
| 350 | 91.2 ± 1.7 | 89.2 ± 2.3 | 74.3 ± 4.4 | 71.6 ± 4.7 |
| 400 | 92.0 ± 2.1 | 89.7 ± 1.8 | 76.3 ± 4.7 | 71.7 ± 6.2 |
| 450 | 94.3 ± 1.7 | 92.0 ± 1.8 | 81.0 ± 6.0 | 74.7 ± 7.0 |
| 500 | 95.0 ± 1.0 | 93.2 ± 1.5 | 80.9 ± 3.9 | 76.5 ± 5.3 |
| 550 | 96.3 ± 0.8 | 94.4 ± 1.1 | 84.9 ± 1.3 | 79.0 ± 3.9 |
| 600 | 97.1 ± 1.3 | 95.3 ± 1.4 | 86.8 ± 4.9 | 80.5 ± 7.0 |
| 650 | 98.3 ± 0.7 | 96.2 ± 0.9 | 90.6 ± 2.9 | 82.3 ± 5.8 |
| 700 | 98.5 ± 0.5 | 97.5 ± 0.5 | 91.9 ± 2.9 | 84.3 ± 5.0 |
| 750 | 99.4 ± 0.3 | 98.1 ± 0.7 | 95.3 ± 2.1 | 87.4 ± 5.1 |
| 800 | 99.7 ± 0.2 | 98.3 ± 0.9 | 96.6 ± 2.2 | 88.4 ± 7.6 |
| 850 | 99.9 ± 0.1 | 99.1 ± 0.5 | 98.1 ± 1.1 | 89.3 ± 7.6 |
| 900 | 100.0 ± 0.0 | 99.2 ± 0.8 | 99.1 ± 0.9 | 91.5 ± 8.9 |
| 950 | 100.0 ± 0.0 | 99.5 ± 0.4 | 99.4 ± 0.6 | 92.6 ± 4.1 |
| 1000 | 100.0 ± 0.0 | 96.5 ± 5.2 | 99.8 ± 0.2 | 83.9 ± 12.7 |
| 1050 | 100.0 ± 0.0 | 90.7 ± 9.4 | 100.0 ± 0.1 | 77.5 ± 17.5 |
| 1100 | 100.0 ± 0.0 | 53.8 ± 33.4 | 99.8 ± 0.3 | 44.3 ± 31.7 |
| 1150 | 100.0 ± 0.0 | 36.2 ± 32.1 | 99.9 ± 0.1 | 32.0 ± 30.3 |
| 1200 | 100.0 ± 0.0 | 20.3 ± 21.7 | 99.8 ± 0.2 | 21.7 ± 23.0 |